build: detect node_g (debug) for available-node in Makefile

[rvagg]

This lets us run test-ci with debug builds, currently we're using a custom invocation of tools/test.py. Since Node 10, we have --mode=$(BUILDTYPE_LOWER) in Makefile which gets us most of the way there. The doc tests in particular rely on available-node to run and since they are part of test-ci we get a failure because it's not looking for the right executable.

Here's a new version of node-test-commit-linux-containered that runs debug using both build-ci then test-ci rather than the current build-ci and custom tools/test.py invocation and it's green for this branch: https://ci.nodejs.org/job/rvagg-test-commit-linux-containered/11/nodes=ubuntu1604_sharedlibs_debug_x64/

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/24540/

[rvagg]

Here's a run that only uses run-ci, via CONFIG_FLAGS=' --debug' make run-ci -j 4: https://ci.nodejs.org/job/rvagg-test-commit-linux-containered/12/nodes=ubuntu1604_sharedlibs_debug_x64

This will partially address nodejs/build#1854, along with #28747 and a new node-test-commit-linux-containered.

[richardlau]

The change looks okay, but I thought passing --debug to configure built both debug and release?

[Trott]

The change looks okay, but I thought passing --debug to configure built both debug and release?

I ran make distclean && ./configure --debug && make -j4 and can confirm that it builds both.

[richardlau]

The change looks okay, but I thought passing --debug to configure built both debug and release?

I ran make distclean && ./configure --debug && make -j4 and can confirm that it builds both.

Which suggests available-node would always find the release node ahead of the debug node_g?

[rvagg]

yeah, there's something odd about this, it fixes what I want fixed but I can't say why, will investigate further

[Trott]

yeah, there's something odd about this, it fixes what I want fixed but I can't say why, will investigate further

Any progress on figuring out why this fixes the issue? Anything I or someone else could look at to help?

[rvagg]

No, haven't had quite enough time to work it out, local debug builds are frustratingly slow!
Current suspicion is that it's a parallelism in make problem, that when it's checking for available-node on the Debug path the Release one isn't available yet, or something like that.

[Trott]

No, haven't had quite enough time to work it out, local debug builds are frustratingly slow!

I seem to recall that either @targos or @tniessen does frequent Debug builds. Maybe they have some insight or else a super-fast machine on which to troubleshoot.

[BridgeAR]

What's the status here?

[rvagg]

status = low confidence, maybe needs to be revisited at a later date or by someone else